Control: notfound -1 5.10.0-36-amd64 . found -1 5.10.244-1 Hello,
5.10.0-36-amd64 is not a kernel version, I guess you meant 5.10.244-1
(i.e. the most recent bullseye kernel) and fixed the bug metadata
accordingly.
On Wed, Dec 10, 2025 at 07:28:32AM -0300, iprodha_gab wrote:
> Package: linux
> Version: 5.10.0-36-amd64
> Severity: important
> Tags: amd64 bullseye disk-performance ahci regression
>
> Problem:
> On AMD A320 chipset systems with AHCI SATA controller, the block I/O layer
> incorrectly displays multiple schedulers as active simultaneously:
> "[mq-deadline] [bfq] none". This causes severe system slowdown and desktop
> unresponsiveness during disk I/O operations.
This is strange.
> How to reproduce:
> 1. Boot Debian 11 (bullseye) on AMD A320 chipset motherboard (ASRock
> A320M-HDV)
> 2. Connect SATA HDD (TOSHIBA DT01ACA0 7200 RPM)
> 3. Check current scheduler: cat /sys/block/sda/queue/scheduler
> 4. Observe output: mq-deadline [bfq] none (note two active schedulers in
> brackets)
Huh, if you repeat the problem description, please make sure the stuff
is consistent.
> 5. Perform large file copy or disk-intensive operation
> 6. Desktop environment (MATE) becomes unresponsive, system stalls
>
> Diagnosis:
> - System: Debian GNU/Linux 11 (bullseye), Kernel: 5.10.0-36-amd64, x86_64
> - Motherboard: ASRock A320M-HDV (AMD Promontory A320 chipset)
> - Disk: TOSHIBA DT01ACA0 (HDD, rotational=1, 7200 RPM SATA)
> - Driver: ahci (multiple AHCI hosts detected: 9 instances)
> - Scheduler state: mq-deadline [bfq] none (bfq active but mq-deadline also
> shown as active)
Another repetition, so you reported a state that looks perfectly fine
twice :-D
> - Expected: Single scheduler in brackets indicating active one
> - Current bug: Two schedulers shown with brackets [mq-deadline] [bfq]
> - Physical system (not virtual machine)
>
> Workaround:
> Temporary fix:
> # echo "bfq" > /sys/block/sda/queue/scheduler
>
> Permanent fix via udev rule:
> ACTION=="add|change", SUBSYSTEM=="block", ATTR{queue/rotational}=="1",
> ATTR{queue/scheduler}="bfq"
>
> Expected behavior:
> According to kernel documentation and block layer design:
> 1. Only ONE scheduler should be active at a time
> 2. The active scheduler should be shown in brackets: [bfq]
> 3. Available schedulers should be listed after: [bfq] mq-deadline none
> 4. Current display "mq-dedeadline [bfq] none" is inconsistent
>
> Impact:
> - Severe desktop usability regression (MATE environment freezes during I/O)
> - System becomes unresponsive for tens of seconds during file operations
> - Affects user experience significantly on HDD-based systems
> - Problem appears to be in scheduler initialization/display logic
>
> Additional notes:
> - Problem persists across reboots
> - Not VM-related (bare metal installation)
> - Multiple AHCI hosts detected (9), may be related to initialization order
> - Manual scheduler selection works correctly once set
>
> Suggested fix locations:
> 1. block/elevator.c - scheduler selection/display logic
> 2. drivers/ata/libata-core.c - ATA device initialization
> 3. /lib/udev/rules.d/60-block-scheduler.rules - default scheduler assignment
>
> System info for reference:
> - Debian: 11 (bullseye)
> - Kernel: 5.10.0-36-amd64
> - Architecture: x86_64
> - Disk: TOSHIBA DT01ACA0 (rotational=1)
> - Driver: ahci (9 instances)
> - Desktop: MATE
> - Motherboard: ASRock A320M-HDV (AMD A320 chipset)
Can you please try if the problem also exists with newer kernels, e.g.
by updating your system to Debian 12 and then Debian 13. Alternatively
at least try a backports kernel.
Best regards
Uwe
signature.asc
Description: PGP signature
